The second session I attended today was hosted by Blaine Cook (formerly of Twitter) and discussed the problems inherent in building the real-time web. His reason for dismissing HTTP as a method for building the real-time web: hard to scale for frequent updates and frequent polling.

I call shenanigans. HTTP is not hard to scale in those situations if you have the right infrastructure. In fact, just about every application delivery controller in existence can easily scale HTTP - even under circumstances described by Blaine. The frequency of updating and polling is similarily a problem with any real-time web application, and in fact could escalate into a problem for highly used real-time applications inside the enterprise as well.

Blaine pointed out other problems with HTTP that I totally agree with as being an impediment to building the real-time web; HTTP doesn't easily support asynchronous communications, there's no standardized API, and time-outs (though I'd argue these are not the fault of HTTP or any other app protocol as much as they are an issue with the underlying delivery system) can cause major hiccups for real-time applications.

Imbibing: Water